Content Machine

Content Machine is a local-first short-form video skill pack for coding agents.

It gives Claude Code, Codex CLI, Cursor, and similar agents the skills, flows, and deterministic runtime tools they need to make TikTok videos, Instagram Reels, and YouTube Shorts without hiding the process.

The point is simple: ask your agent for a video outcome, get back a normal MP4 plus inspectable artifacts for script, audio, timestamps, visuals, captions, render metadata, source rights, and review.

What You Should Understand Fast

• It is for agent harnesses: you already use Claude Code, Codex CLI, Cursor, or a repo-aware coding agent.
• It installs into your repo: the pack materializes .content-machine/skills/ and .content-machine/flows/ so your agent can read them locally.
• It is results-first: Reddit stories over gameplay, motion-card explainers, SaaS shorts, listicles, micro-docs, longform clips, and 3D/procedural gameplay backgrounds.
• It is not a black box: every meaningful step writes files another agent or human can inspect.
• The README is the tour: commands and low-level runtime details are in the deeper docs.

See The Results

The default story example is the familiar Reddit-post-over-gameplay format: full-screen Subway Surfers-style gameplay, a Reddit opener card, then bold captions over the same gameplay.

Repo explainer previews show the same system used for OSS onboarding, motion cards, and 3D/procedural gameplay:

Example	What It Proves	Start Here
demo-9	Reddit story + gameplay is the default low-attention story lane	reddit-post-over-gameplay
demo-18	The repo can explain itself in a social-native Reddit/gameplay format	content-machine-self-demo
demo-19	Deterministic motion cards can explain a repo without stock footage	content-machine-self-demo
demo-20	Code-native 3D/procedural motion can be a rights-safe gameplay background	procedural-gameplay-backgrounds
gallery	The broader lanes: stock explainers, text threads, SaaS, facts, micro-docs, mixed	Showcase Gallery

Use It In Your Repo

Open your agent inside the repo where you want to make videos and paste this:

Install Content Machine in this repo and use it as the local short-form
video skill pack.

Follow the Agent Harness Install guide:
https://github.com/45ck/content-machine/blob/master/docs/user/AGENT-HARNESS-INSTALL.md

Do the install for this harness, materialize `.content-machine/`, write
the right root instruction block for Codex CLI, Claude Code, Cursor, or
this repo-aware agent, then verify the local runner.

After install:
- List the installed skills and flows.
- Run doctor-report.
- Tell me what dependencies, API keys, or media assets are missing.
- For any video request, choose the right Content Machine skill or flow
  first.
- Write artifacts under `runs/<run-id>/`.
- Do not call an MP4 ready unless publish-prep passes.

Install details are in Agent Harness Install. The quick operator path is in Agent Quickstart.

Content Machine does not require a special hosted plugin registry. The agent reads repo-local Markdown skills and flow manifests, then calls the packaged runtime only when execution is needed.

Skills At A Glance

If You Want	Give Your Agent This Skill Or Flow
Full topic-to-video short	generate-short or generate-short.flow
Reddit story over Subway-style gameplay	reddit-post-over-gameplay-short, reddit-card-overlay
Confession or storytime over gameplay	gameplay-confession-short, reddit-story-short
Longform video cut into vertical shorts	longform-to-shorts, longform-clip-extract, reframe-vertical
Product, SaaS, or creator-style ad	saas-problem-solution-short, ugc-avatar-short
Motion cards, diagrams, or coded visuals	motion-card-lesson-short, motion-design-coder
3D/procedural gameplay backgrounds	procedural-gameplay-backgrounds
Hook/title/cover packaging	hook-packaging, platform-packaging, hook-overlay
Virality and retention review	virality-review, retention-pass, publish-prep-review
Captions, timing, and review	short-form-captions, video-render, publish-prep-review
Source media, rights, and reuse	creative-source-scout, source-media-review, asset-ledger
Learn from published results	metrics-feedback-loop, style-profile-library, niche-profile-draft

The complete catalog lives in skills/, executable multi-step paths live in flows/, and deterministic runtime surfaces live in scripts/harness/.

How It Works

Content Machine splits the work so agents do not have to improvise a video pipeline from scratch:

• Skills hold judgment: when to use a lane, what good output looks like, what to reject, and which artifacts should exist.
• Flows coordinate jobs: topic-to-video, longform-to-shorts, reverse-engineering a winner, diagnostics, and repo showcase runs.
• Runtime tools execute stages: script generation, source analysis, audio, timestamps, visuals, captions, render, asset ledger, and publish-prep review.
• Artifacts stay local: runs are meant to be inspectable, resumable, reviewable, and safe for another agent to continue.

What It Can Make

• Reddit/AITA/confession stories over gameplay.
• Stock-footage explainers and faceless mixed-media shorts.
• Motion-card lessons, coded graphics, and 3D/procedural backgrounds.
• Text-message drama, receipt reveals, and social-thread overlays.
• SaaS/product shorts with problem, proof, demo, and CTA beats.
• Fast-facts listicles and micro-documentary breakdowns.
• Longform podcast, interview, talk, or screen-recording clips.
• Reviewed MP4s with provenance, captions, source rights, and quality gates.
• Hook, retention, platform-fit, and metrics feedback loops that improve future shorts without pretending virality is guaranteed.

Start Here

• Use it in your project: Agent Harness Install
• Understand the workflow: Agent Quickstart
• Pick a video lane: Archetypes
• Review hook/retention: Virality And Retention
• Watch examples: Showcase Gallery and Demo Gallery
• Browse skills: Skills
• Browse flows: Flows
• Review readiness: Quality And Review
• Find creative sources: Creative Sources

Repo Map

Path	Why It Exists
skills/	Agent-readable capability docs and examples
flows/	Executable multi-step workflow manifests and operator notes
docs/user/	Human install, quickstart, examples, quality, and showcase docs
docs/demo/	Preview MP4s/GIFs, manifest, provenance, and maturity labels
scripts/harness/	Runtime entrypoints for repo checkouts
agent/	Packaged install and runner binaries
src/harness/	Reusable logic behind agent-facing runtime tools
src/	TypeScript media, render, scoring, validation, and providers
archive/	Frozen legacy CLI/control-plane surfaces

Current Direction

Content Machine is intentionally not a giant all-in-one video app. The active direction is a small, reliable local runtime that coding agents can use well: choose the right short-form lane, preserve intermediate files, review before promotion, and keep the legacy cm shell thin.

Read deeper in Direction, Short-Form Roadmap, and Developer Docs.

Contributing

See CONTRIBUTING.md. Contributor setup, test commands, package smoke checks, and low-level runner examples live in the deeper docs so this README can stay focused on what the project does.

License

MIT